Install Steam
login
|
language
简体中文 (Simplified Chinese)
繁體中文 (Traditional Chinese)
日本語 (Japanese)
한국어 (Korean)
ไทย (Thai)
Български (Bulgarian)
Čeština (Czech)
Dansk (Danish)
Deutsch (German)
Español - España (Spanish - Spain)
Español - Latinoamérica (Spanish - Latin America)
Ελληνικά (Greek)
Français (French)
Italiano (Italian)
Bahasa Indonesia (Indonesian)
Magyar (Hungarian)
Nederlands (Dutch)
Norsk (Norwegian)
Polski (Polish)
Português (Portuguese - Portugal)
Português - Brasil (Portuguese - Brazil)
Română (Romanian)
Русский (Russian)
Suomi (Finnish)
Svenska (Swedish)
Türkçe (Turkish)
Tiếng Việt (Vietnamese)
Українська (Ukrainian)
Report a translation problem
Discussions Rules and Guidelines
Report this post
If you want the best bang for the buck rendering card, I'd look at RX 580's on newegg. I've seen them go for $200
Although, it also depends on the render, depending on what you have in the scene, a GTX 1060 or an RX 480 will be faster, everything affects the rendering time.
For example with my HW: Blenchmark BMW
1. With the i5 3570 processor: 4 min approx
2. With the GTX 660 (old GPU): 3 min approx
3. With the GTX 1060 6GB: 2 min approx
4. With the GTX 1060 + CPU i5 (4 cores): 2 min approx
5. With the GTX 1060 + CPU i5 (1 core): less than 1 min.
* 5 is the best option, but depending on the render it can reduce the time or it may take the same as using only the GPU (BMW is 60% more faster but classrom it takes the same time as using only the GPU).
The RX 480/580 and GTX 1060 3/6 GB are equally at the rendering times.
I'd argue that a Vega 56 at $370 that matches the 1070 in gaming performance but outperforms even a 1080ti in rendering is a much better performance / price than a 1070 that comes in at about $300. Of course power draw is a good factor to keep in mind as Vega will consume quite a lot, but the point is to show that the performance / price doesn't always go to NVIDIA. And of course, this changes when you look at different scenes in blender or when you look at different price points. Just some food for thought. Also, While 580 and 1060 perform the same and cost about the same, the 580 does have 8 gig of VRAM at the same price as 6 from the 1060. Some people may need that.
Sources as of 12/29/2018:
https://www.reddit.com/r/Amd/comments/6tnu9r/vega_64_20_faster_in_blender_than_gtx_1080ti/
https://www.newegg.com/Product/Product.aspx?Item=N82E16814131740&Description=Vega%2056&cm_re=Vega_56-_-14-131-740-_-Product
https://www.newegg.com/Product/Product.aspx?Item=N82E16814126119&Description=1070&cm_re=1070-_-14-126-119-_-Product
What is going on here?
Also, can I ask what are the times for the following...
With the GTX 1060 + CPU i5 (3 cores): ?
With the GTX 1060 + CPU i5 (2 cores): ?
Thanks in advance.
I’m not 100% sure about what the times are for, but I believe it has to do with rendering with the CPU and GPU and how combined rendering like that in blender isn’t that good.
As for my response with the links, you don’t really have to pay attention to that.
When you use GPU render without CPUc your CPU isn’t gonna do much in the way of changing the render time.
While I heavily recommend getting a new CPU, that would entail getting a new motherboard and RAM, and wouldn’t do too much for render times.
I’d you’re looking for a cheap very good price to performance card, a GTX 1060 6 gig (the 6 gig makes a difference as it has more cuda cores and 3 gigs of vram can be limiting) or an RX 580 will probably be your best choice. Both can be had for $200 on Newegg.
If you have another price range in mind, we can help figure you out a good path for upgrading, as you can plan ahead for a future computer with what you buy now.
I wasn't the OP but wanted to ask what I did as it is interesting the difference using different number of CPU cores so wanted to see if Mika would mind checking the BMW test for gfx card + 3 cores and then gfx card + 2 cores to see what the times were.
A redit post, that is not correct. The vega 64 gives a performance slightly greater than the GTX 1070 in both games and Blender. If you look for information in all cases the VEGA people put it behind the GTX 1080. You must take into account that OpenCL does not give all the performance of the AMD and also depends on the rest of HW. I use a GTX 1060 and I do not get all the performance for using a motherboard with a cheap chipset. Apart there is also the detail of the drivers, both in AMD and Nvidia the more you update the driver the worse the performance goes. AMD does not show much, but Nvidia does. In terms of having 6GB or 8GB to date I have not exceeded 3GB of VRAM in any game or blender even putting the games in 4K Ultra.
The price, in Spain depends on the time it costs the same one RX580 as a GTX 1060 or it can cost a different price. Right now the difference is € 20 to € 50.
The time is less than using 4 cores but greater than using 1.
The reason is very simple. As SEAN say, Blender is not very efficient in this matter.
The first core of the processor is attached to the GPU which reduces the time considerably, however the cores 2, 3 and 4 don't join to the GPU and go separately, which will take the same time as the processor without using the GPU .
For example. The GPU takes 4 seconds to calculate a tile. Adding a processor core reduces the time to 3 seconds. Now, if the processor core goes alone then it will take 35 seconds per tile / core. In this case the GPU will calculate all the tiles in 3 seconds, if there are 24 tiles, 3 cores will be left with 6 tiles and the GPU with 18 tiles, the result is 54 seconds for GPU, but of course the remaining cores will take 35 seconds each then that's it will be 70 seconds, then the GPU and first core will take less time and then you will have to wait for the rest of the cores to finish. If you reduce the number of cores the result is smaller because the GPU and first core render more tiles, but you will still have to wait for the second and third cores to finish. That's why the best option is to use a single processor core.
What they must do (blender team) is that all the cores work next to the GPU. This in Cinema 4D works correctly since you can join the cores in a single tile although in this case it takes the same time to render a tile with 4 cores than 4 tiles with 4 cores. Cinema 4D doesn't have to work with GPU but it can work with other PCs and use all the cores it has.
But are you adjusting the tile/bucket size?
Because it's faster to render Cpu using a small bucket size.. and large tile size for GPU.
I did read the thread... but -- you did ask for other suggestions.
Here's a little chart.. for reference:
https://www.google.com/imgres?imgurl=https%3A%2F%2Fstatic1.squarespace.com%2Fstatic%2F58586fa5ebbd1a60e7d76d3e%2Ft%2F591058e2d2b8578e647eabc7%2F1494243557118%2F&imgrefurl=https%3A%2F%2Fwww.blenderguru.com%2Farticles%2F4-easy-ways-to-speed-up-cycles&docid=GcXxgt-YCHI-rM&tbnid=zjZkyzb4xjxbOM%3A&vet=10ahUKEwjc9cGV5sffAhWhuXEKHZ98A5IQMwg-KAEwAQ..i&w=495&h=310&bih=741&biw=972&q=blender%20tile%20size&ved=0ahUKEwjc9cGV5sffAhWhuXEKHZ98A5IQMwg-KAEwAQ&iact=mrc&uact=8
Yeah, I did that. Reduced the tile to 16x16, reduced samples to around a hundred, two hundred, disabled caustics (my scene has a floor with reflection). All of that.
Still, it surprises me that I can play for example TF2 who has more particles going on with the flamethrower and unusuals together than rendering a mere 400 particles scene. The particles are the only thing that moves in my scene.
With all those changes the render time for each frame went down from one minute and a half to forty seconds. But the whole scene takes five hours to render.
I would like to reduce it to atleast 20 seconds per frame. That's why I wanted to see if I had any luck turning on OpenCL.
You cannot compare particles in TF2 to particles in Blender (or generally "rendering" in a game engine to rendering with a real render engine). Two completely different things.
What would be the difference? I have never thinkered with game engines but I assume they work sort of the same when comes to rendering. After all, movement inside video, games and pictures is overlapping the next frame at certain speed.
cant speak for -every- engine but they typically don't rebake the entire lighting solution on a frame-by-frame basis.. as blender is doing, when yuo just stick everything in a scene and hit 'render'
When you're rendering in blender you're basically calculating the lighting for anything and everything in the scene for every frame, you can also pre-bake the lighting solution in blender to save on rendering for animation.. for the 'background' for example, and just render your animated particles... then composite the two frames together.. thus saving on render time because you're only rendering the things that are moving...
but ill let you google how to do that because it's a fairly involved process... but i have just explained the 'general theory' there..
Ah, I get it now. Didn't knew it worked that way. I was searching around about that but didn't find much right away. Thanks.
(i overheard that that Linux is better for Blender)
My CPU is faster then the GPU when it comes to render a picture in Blender.
Your rendering doesnt Hang your system and you can work on something diffrent even in Blender.
And by using Cpus you have the benift that blender takes then whole MAIN-RAM
And the best comes to the end: it never crashes during render.
because by using your gpu you are limited to the amount of ram installed on your gfx-card
Case of "gruengruen" sure, the GTX 960 is low power so you will have better results with the processor. If the graph was a GTX 980 / RX470 or higher then the processor would be slower.
The issue of RAM, BLENDER is not able to use virtual memory so that whether VRAM or RAM is consumed at 100% blender stops working. The graphics card has less memory normally than RAM however if you are able to optimize your scene you should not have problems. I didn't know how to optimize and I consume up to 4 - 6 GB of RAM or VRAM, which I was short with a GTX 660 2GB, then I buy a GTX 1060 6GB and learned to optimize and to date I have not exceed 3GB of VRAM in the cases that consume more memory and with a very large render. Most cases I haven't exceeded 1GB of VRAM despite the complexity of the scene. Almost always it has been for arrays of more than 1000 or emitters in large quantities.
Linux, yes, can accelerate up to 40% of the render time, however only by CPU, if is the GPU doesn't improve the time.
On the other hand the GPU doesn't stuck the system, now it does that the image update is slow but that doesn't stop me from playing other games of Steam like Clicker or idle (for example). In case of using the CPU that if it blocks the whole computer and therefore I can't use anything while using the entire processor. If I put render by CPU + GPU (with a single core) from the task manager change so that blender only use the core 2, 3 or 4, this way the core 1 is unused and I can take advantage to do other things without problems.